RESEARCH: New shared AI for smart buildings optimizes their management, improves data privacy

Researchers at the Cybersecurity and Intelligent Systems Engineering have developed a new artificial intelligence (AI) system that could make smart buildings more efficient, private and easier to manage. The framework, called a “federated neuro-symbolic rule learning,” is the first of its kind designed specifically for building operations.

Smart buildings rely on sensors to track things like occupancy, activities and energy use to automate heating, lighting and appliances. But collecting such data raises privacy concerns and often requires powerful, resource-heavy AI systems that are difficult to run on small devices.

The new approach addresses both problems. Instead of sending raw data to a central server, it uses federated learning, where multiple buildings train a shared model while keeping their data local.

At the same time, the technology combines neural networks with symbolic reasoning to generate simple, human-readable “if-then” rules. These rules can run directly on low-power devices and explain how decisions are made.

For example, the system can learn rules such as: If motion is detected in the kitchen in the evening, cooking is likely happening. This makes the AI more transparent and useful for real-world decision-making.

Tests show the model outperforms both traditional rule-based systems and modern deep learning approaches. It achieved up to 45 per cent higher accuracy, while being significantly smaller and up to 200 times faster than some deep learning models. It also performed better on new, unseen data, improving generalization by about 20 per cent.

By combining privacy protection, efficiency and explainability, this research offers a practical path toward smarter, more trustworthy building automation systems.

The study was authored by Fatimah Faiza Farrukh and Manar Amayri, and was published in the journal Energy and Buildings. The research was supported by the Natural Sciences and Engineering Research Council of Canada.

Read the cited paper: “Federated neuro-symbolic rule-learning for lightweight smart building operations”